“It will never, ever say no, even if it’s incapable, even if it has no thoughts, even if what you are asking is equal parts impossible and unreasonable in both its timescale and scope.”
Our reading:
LLM coding tools do not refuse requests even when the request is impossible or beyond their ability.
Assessment: Unresolved
What happened
Was it true for the period described?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
A preprint published September 3, 2026 reports a 27% aggregate refusal rate on adversarial coding prompts across its tested models [e1]. It describes tasks where the correct response is to refuse or flag impossibility [e1]. The supplied record does not date those interactions to May 1–26, 2026 or establish the behavior of Claude Code or Cursor during that period [e1]. The later results therefore do not settle the contracted May observation [e1].
Caveats
- The reported rate applies to the study's tested models and prompts, not directly to the deployed coding tools named in the passage [e1].
- The study supplies no authenticated interaction dated within the assessed May period [e1].
Evidence (1 item)
Context only: arXiv; Vishnu Asutosh Dasu, Ashish Kundu and Gang Tan,
RR is 0.270.27.
Why it matters: The paper reports a 27% refusal rate on deliberately unsatisfiable coding prompts. This challenges a timeless categorical reading, but the retrieved record does not establish that these interactions occurred during May 2026.
- Publisher
- arXiv; Vishnu Asutosh Dasu, Ashish Kundu and Gang Tan
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Where we had to interpret
- Read as a categorical description of contemporary LLM coding tools, including Claude Code and Cursor, rather than a forecast about all future models.
- Saying no includes explicitly acknowledging that the requested work cannot be completed under the stated constraints. Safety-policy refusals, permission errors and service failures do not establish this behavior.
- The ongoing observation is assessed during May 2026 through publication. Software behavior can be observed immediately; it does not require a quarterly reporting cycle.
Right if: Representative contemporary observations consistently show tools accepting infeasible coding requests without acknowledging their inability. A finite sample can support a tendency but cannot establish the literal universal.
Wrong if: An authenticated interaction or adequately documented evaluation within the target period shows a coding tool explicitly declining or abandoning a request because it is impossible or beyond its capabilities.
We wrote this test after the outcome was known.
What would settle it
An authenticated May 1–26, 2026 interaction or adequately documented evaluation showing a coding tool explicitly declining or abandoning an infeasible request would contradict the categorical claim. The record must include the model, tool version, date, prompt, relevant instructions and complete response, and distinguish infeasibility from safety refusals, permission errors or service failures. A finite sample without refusals could support a tendency but could not prove 'never.'
We found 42 claims in this post. See them all.
Full record
Source and dates
Where's Your Ed At, Revenge of The Business Idiot, paragraph 13.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-28. The post may have been edited after 2026-05-26.
Where it was said (1)
Said once in the posts we read.
: Where's Your Ed At, Revenge of The Business Idiot
It will never, ever say no, even if it’s incapable, even if it has no thoughts, even if what you are asking is equal parts impossible and unreasonable in both its timescale and scope.
- Where
- paragraph 13
- Published
- 2026-05-26
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.
Held backBy us
The final check failed: the write-up said more than the evidence showed.